xref: /linux/arch/arm64/kernel/head.S (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1/*
2 * Low-level CPU initialisation
3 * Based on arch/arm/kernel/head.S
4 *
5 * Copyright (C) 1994-2002 Russell King
6 * Copyright (C) 2003-2012 ARM Ltd.
7 * Authors:	Catalin Marinas <catalin.marinas@arm.com>
8 *		Will Deacon <will.deacon@arm.com>
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17 * GNU General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21 */
22
23#include <linux/linkage.h>
24#include <linux/init.h>
25#include <linux/irqchip/arm-gic-v3.h>
26
27#include <asm/assembler.h>
28#include <asm/boot.h>
29#include <asm/ptrace.h>
30#include <asm/asm-offsets.h>
31#include <asm/cache.h>
32#include <asm/cputype.h>
33#include <asm/elf.h>
34#include <asm/kernel-pgtable.h>
35#include <asm/kvm_arm.h>
36#include <asm/memory.h>
37#include <asm/pgtable-hwdef.h>
38#include <asm/pgtable.h>
39#include <asm/page.h>
40#include <asm/smp.h>
41#include <asm/sysreg.h>
42#include <asm/thread_info.h>
43#include <asm/virt.h>
44
45#define __PHYS_OFFSET	(KERNEL_START - TEXT_OFFSET)
46
47#if (TEXT_OFFSET & 0xfff) != 0
48#error TEXT_OFFSET must be at least 4KB aligned
49#elif (PAGE_OFFSET & 0x1fffff) != 0
50#error PAGE_OFFSET must be at least 2MB aligned
51#elif TEXT_OFFSET > 0x1fffff
52#error TEXT_OFFSET must be less than 2MB
53#endif
54
55/*
56 * Kernel startup entry point.
57 * ---------------------------
58 *
59 * The requirements are:
60 *   MMU = off, D-cache = off, I-cache = on or off,
61 *   x0 = physical address to the FDT blob.
62 *
63 * This code is mostly position independent so you call this at
64 * __pa(PAGE_OFFSET + TEXT_OFFSET).
65 *
66 * Note that the callee-saved registers are used for storing variables
67 * that are useful before the MMU is enabled. The allocations are described
68 * in the entry routines.
69 */
70	__HEAD
71_head:
72	/*
73	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
74	 */
75#ifdef CONFIG_EFI
76	/*
77	 * This add instruction has no meaningful effect except that
78	 * its opcode forms the magic "MZ" signature required by UEFI.
79	 */
80	add	x13, x18, #0x16
81	b	stext
82#else
83	b	stext				// branch to kernel start, magic
84	.long	0				// reserved
85#endif
86	le64sym	_kernel_offset_le		// Image load offset from start of RAM, little-endian
87	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
88	le64sym	_kernel_flags_le		// Informative flags, little-endian
89	.quad	0				// reserved
90	.quad	0				// reserved
91	.quad	0				// reserved
92	.byte	0x41				// Magic number, "ARM\x64"
93	.byte	0x52
94	.byte	0x4d
95	.byte	0x64
96#ifdef CONFIG_EFI
97	.long	pe_header - _head		// Offset to the PE header.
98#else
99	.word	0				// reserved
100#endif
101
102#ifdef CONFIG_EFI
103	.align 3
104pe_header:
105	.ascii	"PE"
106	.short 	0
107coff_header:
108	.short	0xaa64				// AArch64
109	.short	2				// nr_sections
110	.long	0 				// TimeDateStamp
111	.long	0				// PointerToSymbolTable
112	.long	1				// NumberOfSymbols
113	.short	section_table - optional_header	// SizeOfOptionalHeader
114	.short	0x206				// Characteristics.
115						// IMAGE_FILE_DEBUG_STRIPPED |
116						// IMAGE_FILE_EXECUTABLE_IMAGE |
117						// IMAGE_FILE_LINE_NUMS_STRIPPED
118optional_header:
119	.short	0x20b				// PE32+ format
120	.byte	0x02				// MajorLinkerVersion
121	.byte	0x14				// MinorLinkerVersion
122	.long	_end - efi_header_end		// SizeOfCode
123	.long	0				// SizeOfInitializedData
124	.long	0				// SizeOfUninitializedData
125	.long	__efistub_entry - _head		// AddressOfEntryPoint
126	.long	efi_header_end - _head		// BaseOfCode
127
128extra_header_fields:
129	.quad	0				// ImageBase
130	.long	0x1000				// SectionAlignment
131	.long	PECOFF_FILE_ALIGNMENT		// FileAlignment
132	.short	0				// MajorOperatingSystemVersion
133	.short	0				// MinorOperatingSystemVersion
134	.short	0				// MajorImageVersion
135	.short	0				// MinorImageVersion
136	.short	0				// MajorSubsystemVersion
137	.short	0				// MinorSubsystemVersion
138	.long	0				// Win32VersionValue
139
140	.long	_end - _head			// SizeOfImage
141
142	// Everything before the kernel image is considered part of the header
143	.long	efi_header_end - _head		// SizeOfHeaders
144	.long	0				// CheckSum
145	.short	0xa				// Subsystem (EFI application)
146	.short	0				// DllCharacteristics
147	.quad	0				// SizeOfStackReserve
148	.quad	0				// SizeOfStackCommit
149	.quad	0				// SizeOfHeapReserve
150	.quad	0				// SizeOfHeapCommit
151	.long	0				// LoaderFlags
152	.long	(section_table - .) / 8		// NumberOfRvaAndSizes
153
154	.quad	0				// ExportTable
155	.quad	0				// ImportTable
156	.quad	0				// ResourceTable
157	.quad	0				// ExceptionTable
158	.quad	0				// CertificationTable
159	.quad	0				// BaseRelocationTable
160
161#ifdef CONFIG_DEBUG_EFI
162	.long	efi_debug_table - _head		// DebugTable
163	.long	efi_debug_table_size
164#endif
165
166	// Section table
167section_table:
168
169	/*
170	 * The EFI application loader requires a relocation section
171	 * because EFI applications must be relocatable.  This is a
172	 * dummy section as far as we are concerned.
173	 */
174	.ascii	".reloc"
175	.byte	0
176	.byte	0			// end of 0 padding of section name
177	.long	0
178	.long	0
179	.long	0			// SizeOfRawData
180	.long	0			// PointerToRawData
181	.long	0			// PointerToRelocations
182	.long	0			// PointerToLineNumbers
183	.short	0			// NumberOfRelocations
184	.short	0			// NumberOfLineNumbers
185	.long	0x42100040		// Characteristics (section flags)
186
187
188	.ascii	".text"
189	.byte	0
190	.byte	0
191	.byte	0        		// end of 0 padding of section name
192	.long	_end - efi_header_end	// VirtualSize
193	.long	efi_header_end - _head	// VirtualAddress
194	.long	_edata - efi_header_end	// SizeOfRawData
195	.long	efi_header_end - _head	// PointerToRawData
196
197	.long	0		// PointerToRelocations (0 for executables)
198	.long	0		// PointerToLineNumbers (0 for executables)
199	.short	0		// NumberOfRelocations  (0 for executables)
200	.short	0		// NumberOfLineNumbers  (0 for executables)
201	.long	0xe0500020	// Characteristics (section flags)
202
203#ifdef CONFIG_DEBUG_EFI
204	/*
205	 * The debug table is referenced via its Relative Virtual Address (RVA),
206	 * which is only defined for those parts of the image that are covered
207	 * by a section declaration. Since this header is not covered by any
208	 * section, the debug table must be emitted elsewhere. So stick it in
209	 * the .init.rodata section instead.
210	 *
211	 * Note that the EFI debug entry itself may legally have a zero RVA,
212	 * which means we can simply put it right after the section headers.
213	 */
214	__INITRODATA
215
216	.align	2
217efi_debug_table:
218	// EFI_IMAGE_DEBUG_DIRECTORY_ENTRY
219	.long	0			// Characteristics
220	.long	0			// TimeDateStamp
221	.short	0			// MajorVersion
222	.short	0			// MinorVersion
223	.long	2			// Type == EFI_IMAGE_DEBUG_TYPE_CODEVIEW
224	.long	efi_debug_entry_size	// SizeOfData
225	.long	0			// RVA
226	.long	efi_debug_entry - _head	// FileOffset
227
228	.set	efi_debug_table_size, . - efi_debug_table
229	.previous
230
231efi_debug_entry:
232	// EFI_IMAGE_DEBUG_CODEVIEW_NB10_ENTRY
233	.ascii	"NB10"			// Signature
234	.long	0			// Unknown
235	.long	0			// Unknown2
236	.long	0			// Unknown3
237
238	.asciz	VMLINUX_PATH
239
240	.set	efi_debug_entry_size, . - efi_debug_entry
241#endif
242
243	/*
244	 * EFI will load .text onwards at the 4k section alignment
245	 * described in the PE/COFF header. To ensure that instruction
246	 * sequences using an adrp and a :lo12: immediate will function
247	 * correctly at this alignment, we must ensure that .text is
248	 * placed at a 4k boundary in the Image to begin with.
249	 */
250	.align 12
251efi_header_end:
252#endif
253
254	__INIT
255
256	/*
257	 * The following callee saved general purpose registers are used on the
258	 * primary lowlevel boot path:
259	 *
260	 *  Register   Scope                      Purpose
261	 *  x21        stext() .. start_kernel()  FDT pointer passed at boot in x0
262	 *  x23        stext() .. start_kernel()  physical misalignment/KASLR offset
263	 *  x28        __create_page_tables()     callee preserved temp register
264	 *  x19/x20    __primary_switch()         callee preserved temp registers
265	 */
266ENTRY(stext)
267	bl	preserve_boot_args
268	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
269	adrp	x23, __PHYS_OFFSET
270	and	x23, x23, MIN_KIMG_ALIGN - 1	// KASLR offset, defaults to 0
271	bl	set_cpu_boot_mode_flag
272	bl	__create_page_tables
273	/*
274	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
275	 * details.
276	 * On return, the CPU will be ready for the MMU to be turned on and
277	 * the TCR will have been set.
278	 */
279	bl	__cpu_setup			// initialise processor
280	b	__primary_switch
281ENDPROC(stext)
282
283/*
284 * Preserve the arguments passed by the bootloader in x0 .. x3
285 */
286preserve_boot_args:
287	mov	x21, x0				// x21=FDT
288
289	adr_l	x0, boot_args			// record the contents of
290	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
291	stp	x2, x3, [x0, #16]
292
293	dmb	sy				// needed before dc ivac with
294						// MMU off
295
296	add	x1, x0, #0x20			// 4 x 8 bytes
297	b	__inval_cache_range		// tail call
298ENDPROC(preserve_boot_args)
299
300/*
301 * Macro to create a table entry to the next page.
302 *
303 *	tbl:	page table address
304 *	virt:	virtual address
305 *	shift:	#imm page table shift
306 *	ptrs:	#imm pointers per table page
307 *
308 * Preserves:	virt
309 * Corrupts:	tmp1, tmp2
310 * Returns:	tbl -> next level table page address
311 */
312	.macro	create_table_entry, tbl, virt, shift, ptrs, tmp1, tmp2
313	lsr	\tmp1, \virt, #\shift
314	and	\tmp1, \tmp1, #\ptrs - 1	// table index
315	add	\tmp2, \tbl, #PAGE_SIZE
316	orr	\tmp2, \tmp2, #PMD_TYPE_TABLE	// address of next table and entry type
317	str	\tmp2, [\tbl, \tmp1, lsl #3]
318	add	\tbl, \tbl, #PAGE_SIZE		// next level table page
319	.endm
320
321/*
322 * Macro to populate the PGD (and possibily PUD) for the corresponding
323 * block entry in the next level (tbl) for the given virtual address.
324 *
325 * Preserves:	tbl, next, virt
326 * Corrupts:	tmp1, tmp2
327 */
328	.macro	create_pgd_entry, tbl, virt, tmp1, tmp2
329	create_table_entry \tbl, \virt, PGDIR_SHIFT, PTRS_PER_PGD, \tmp1, \tmp2
330#if SWAPPER_PGTABLE_LEVELS > 3
331	create_table_entry \tbl, \virt, PUD_SHIFT, PTRS_PER_PUD, \tmp1, \tmp2
332#endif
333#if SWAPPER_PGTABLE_LEVELS > 2
334	create_table_entry \tbl, \virt, SWAPPER_TABLE_SHIFT, PTRS_PER_PTE, \tmp1, \tmp2
335#endif
336	.endm
337
338/*
339 * Macro to populate block entries in the page table for the start..end
340 * virtual range (inclusive).
341 *
342 * Preserves:	tbl, flags
343 * Corrupts:	phys, start, end, pstate
344 */
345	.macro	create_block_map, tbl, flags, phys, start, end
346	lsr	\phys, \phys, #SWAPPER_BLOCK_SHIFT
347	lsr	\start, \start, #SWAPPER_BLOCK_SHIFT
348	and	\start, \start, #PTRS_PER_PTE - 1	// table index
349	orr	\phys, \flags, \phys, lsl #SWAPPER_BLOCK_SHIFT	// table entry
350	lsr	\end, \end, #SWAPPER_BLOCK_SHIFT
351	and	\end, \end, #PTRS_PER_PTE - 1		// table end index
3529999:	str	\phys, [\tbl, \start, lsl #3]		// store the entry
353	add	\start, \start, #1			// next entry
354	add	\phys, \phys, #SWAPPER_BLOCK_SIZE		// next block
355	cmp	\start, \end
356	b.ls	9999b
357	.endm
358
359/*
360 * Setup the initial page tables. We only setup the barest amount which is
361 * required to get the kernel running. The following sections are required:
362 *   - identity mapping to enable the MMU (low address, TTBR0)
363 *   - first few MB of the kernel linear mapping to jump to once the MMU has
364 *     been enabled
365 */
366__create_page_tables:
367	mov	x28, lr
368
369	/*
370	 * Invalidate the idmap and swapper page tables to avoid potential
371	 * dirty cache lines being evicted.
372	 */
373	adrp	x0, idmap_pg_dir
374	adrp	x1, swapper_pg_dir + SWAPPER_DIR_SIZE + RESERVED_TTBR0_SIZE
375	bl	__inval_cache_range
376
377	/*
378	 * Clear the idmap and swapper page tables.
379	 */
380	adrp	x0, idmap_pg_dir
381	adrp	x6, swapper_pg_dir + SWAPPER_DIR_SIZE + RESERVED_TTBR0_SIZE
3821:	stp	xzr, xzr, [x0], #16
383	stp	xzr, xzr, [x0], #16
384	stp	xzr, xzr, [x0], #16
385	stp	xzr, xzr, [x0], #16
386	cmp	x0, x6
387	b.lo	1b
388
389	mov	x7, SWAPPER_MM_MMUFLAGS
390
391	/*
392	 * Create the identity mapping.
393	 */
394	adrp	x0, idmap_pg_dir
395	adrp	x3, __idmap_text_start		// __pa(__idmap_text_start)
396
397#ifndef CONFIG_ARM64_VA_BITS_48
398#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
399#define EXTRA_PTRS	(1 << (48 - EXTRA_SHIFT))
400
401	/*
402	 * If VA_BITS < 48, it may be too small to allow for an ID mapping to be
403	 * created that covers system RAM if that is located sufficiently high
404	 * in the physical address space. So for the ID map, use an extended
405	 * virtual range in that case, by configuring an additional translation
406	 * level.
407	 * First, we have to verify our assumption that the current value of
408	 * VA_BITS was chosen such that all translation levels are fully
409	 * utilised, and that lowering T0SZ will always result in an additional
410	 * translation level to be configured.
411	 */
412#if VA_BITS != EXTRA_SHIFT
413#error "Mismatch between VA_BITS and page size/number of translation levels"
414#endif
415
416	/*
417	 * Calculate the maximum allowed value for TCR_EL1.T0SZ so that the
418	 * entire ID map region can be mapped. As T0SZ == (64 - #bits used),
419	 * this number conveniently equals the number of leading zeroes in
420	 * the physical address of __idmap_text_end.
421	 */
422	adrp	x5, __idmap_text_end
423	clz	x5, x5
424	cmp	x5, TCR_T0SZ(VA_BITS)	// default T0SZ small enough?
425	b.ge	1f			// .. then skip additional level
426
427	adr_l	x6, idmap_t0sz
428	str	x5, [x6]
429	dmb	sy
430	dc	ivac, x6		// Invalidate potentially stale cache line
431
432	create_table_entry x0, x3, EXTRA_SHIFT, EXTRA_PTRS, x5, x6
4331:
434#endif
435
436	create_pgd_entry x0, x3, x5, x6
437	mov	x5, x3				// __pa(__idmap_text_start)
438	adr_l	x6, __idmap_text_end		// __pa(__idmap_text_end)
439	create_block_map x0, x7, x3, x5, x6
440
441	/*
442	 * Map the kernel image (starting with PHYS_OFFSET).
443	 */
444	adrp	x0, swapper_pg_dir
445	mov_q	x5, KIMAGE_VADDR + TEXT_OFFSET	// compile time __va(_text)
446	add	x5, x5, x23			// add KASLR displacement
447	create_pgd_entry x0, x5, x3, x6
448	adrp	x6, _end			// runtime __pa(_end)
449	adrp	x3, _text			// runtime __pa(_text)
450	sub	x6, x6, x3			// _end - _text
451	add	x6, x6, x5			// runtime __va(_end)
452	create_block_map x0, x7, x3, x5, x6
453
454	/*
455	 * Since the page tables have been populated with non-cacheable
456	 * accesses (MMU disabled), invalidate the idmap and swapper page
457	 * tables again to remove any speculatively loaded cache lines.
458	 */
459	adrp	x0, idmap_pg_dir
460	adrp	x1, swapper_pg_dir + SWAPPER_DIR_SIZE + RESERVED_TTBR0_SIZE
461	dmb	sy
462	bl	__inval_cache_range
463
464	ret	x28
465ENDPROC(__create_page_tables)
466	.ltorg
467
468/*
469 * The following fragment of code is executed with the MMU enabled.
470 *
471 *   x0 = __PHYS_OFFSET
472 */
473__primary_switched:
474	adrp	x4, init_thread_union
475	add	sp, x4, #THREAD_SIZE
476	adr_l	x5, init_task
477	msr	sp_el0, x5			// Save thread_info
478
479	adr_l	x8, vectors			// load VBAR_EL1 with virtual
480	msr	vbar_el1, x8			// vector table address
481	isb
482
483	stp	xzr, x30, [sp, #-16]!
484	mov	x29, sp
485
486	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
487
488	ldr_l	x4, kimage_vaddr		// Save the offset between
489	sub	x4, x4, x0			// the kernel virtual and
490	str_l	x4, kimage_voffset, x5		// physical mappings
491
492	// Clear BSS
493	adr_l	x0, __bss_start
494	mov	x1, xzr
495	adr_l	x2, __bss_stop
496	sub	x2, x2, x0
497	bl	__pi_memset
498	dsb	ishst				// Make zero page visible to PTW
499
500#ifdef CONFIG_KASAN
501	bl	kasan_early_init
502#endif
503#ifdef CONFIG_RANDOMIZE_BASE
504	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
505	b.ne	0f
506	mov	x0, x21				// pass FDT address in x0
507	mov	x1, x23				// pass modulo offset in x1
508	bl	kaslr_early_init		// parse FDT for KASLR options
509	cbz	x0, 0f				// KASLR disabled? just proceed
510	orr	x23, x23, x0			// record KASLR offset
511	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
512	ret					// to __primary_switch()
5130:
514#endif
515	b	start_kernel
516ENDPROC(__primary_switched)
517
518/*
519 * end early head section, begin head code that is also used for
520 * hotplug and needs to have the same protections as the text region
521 */
522	.section ".idmap.text","ax"
523
524ENTRY(kimage_vaddr)
525	.quad		_text - TEXT_OFFSET
526
527/*
528 * If we're fortunate enough to boot at EL2, ensure that the world is
529 * sane before dropping to EL1.
530 *
531 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
532 * booted in EL1 or EL2 respectively.
533 */
534ENTRY(el2_setup)
535	mrs	x0, CurrentEL
536	cmp	x0, #CurrentEL_EL2
537	b.ne	1f
538	mrs	x0, sctlr_el2
539CPU_BE(	orr	x0, x0, #(1 << 25)	)	// Set the EE bit for EL2
540CPU_LE(	bic	x0, x0, #(1 << 25)	)	// Clear the EE bit for EL2
541	msr	sctlr_el2, x0
542	b	2f
5431:	mrs	x0, sctlr_el1
544CPU_BE(	orr	x0, x0, #(3 << 24)	)	// Set the EE and E0E bits for EL1
545CPU_LE(	bic	x0, x0, #(3 << 24)	)	// Clear the EE and E0E bits for EL1
546	msr	sctlr_el1, x0
547	mov	w0, #BOOT_CPU_MODE_EL1		// This cpu booted in EL1
548	isb
549	ret
550
5512:
552#ifdef CONFIG_ARM64_VHE
553	/*
554	 * Check for VHE being present. For the rest of the EL2 setup,
555	 * x2 being non-zero indicates that we do have VHE, and that the
556	 * kernel is intended to run at EL2.
557	 */
558	mrs	x2, id_aa64mmfr1_el1
559	ubfx	x2, x2, #8, #4
560#else
561	mov	x2, xzr
562#endif
563
564	/* Hyp configuration. */
565	mov	x0, #HCR_RW			// 64-bit EL1
566	cbz	x2, set_hcr
567	orr	x0, x0, #HCR_TGE		// Enable Host Extensions
568	orr	x0, x0, #HCR_E2H
569set_hcr:
570	msr	hcr_el2, x0
571	isb
572
573	/*
574	 * Allow Non-secure EL1 and EL0 to access physical timer and counter.
575	 * This is not necessary for VHE, since the host kernel runs in EL2,
576	 * and EL0 accesses are configured in the later stage of boot process.
577	 * Note that when HCR_EL2.E2H == 1, CNTHCTL_EL2 has the same bit layout
578	 * as CNTKCTL_EL1, and CNTKCTL_EL1 accessing instructions are redefined
579	 * to access CNTHCTL_EL2. This allows the kernel designed to run at EL1
580	 * to transparently mess with the EL0 bits via CNTKCTL_EL1 access in
581	 * EL2.
582	 */
583	cbnz	x2, 1f
584	mrs	x0, cnthctl_el2
585	orr	x0, x0, #3			// Enable EL1 physical timers
586	msr	cnthctl_el2, x0
5871:
588	msr	cntvoff_el2, xzr		// Clear virtual offset
589
590#ifdef CONFIG_ARM_GIC_V3
591	/* GICv3 system register access */
592	mrs	x0, id_aa64pfr0_el1
593	ubfx	x0, x0, #24, #4
594	cmp	x0, #1
595	b.ne	3f
596
597	mrs_s	x0, ICC_SRE_EL2
598	orr	x0, x0, #ICC_SRE_EL2_SRE	// Set ICC_SRE_EL2.SRE==1
599	orr	x0, x0, #ICC_SRE_EL2_ENABLE	// Set ICC_SRE_EL2.Enable==1
600	msr_s	ICC_SRE_EL2, x0
601	isb					// Make sure SRE is now set
602	mrs_s	x0, ICC_SRE_EL2			// Read SRE back,
603	tbz	x0, #0, 3f			// and check that it sticks
604	msr_s	ICH_HCR_EL2, xzr		// Reset ICC_HCR_EL2 to defaults
605
6063:
607#endif
608
609	/* Populate ID registers. */
610	mrs	x0, midr_el1
611	mrs	x1, mpidr_el1
612	msr	vpidr_el2, x0
613	msr	vmpidr_el2, x1
614
615	/*
616	 * When VHE is not in use, early init of EL2 and EL1 needs to be
617	 * done here.
618	 * When VHE _is_ in use, EL1 will not be used in the host and
619	 * requires no configuration, and all non-hyp-specific EL2 setup
620	 * will be done via the _EL1 system register aliases in __cpu_setup.
621	 */
622	cbnz	x2, 1f
623
624	/* sctlr_el1 */
625	mov	x0, #0x0800			// Set/clear RES{1,0} bits
626CPU_BE(	movk	x0, #0x33d0, lsl #16	)	// Set EE and E0E on BE systems
627CPU_LE(	movk	x0, #0x30d0, lsl #16	)	// Clear EE and E0E on LE systems
628	msr	sctlr_el1, x0
629
630	/* Coprocessor traps. */
631	mov	x0, #0x33ff
632	msr	cptr_el2, x0			// Disable copro. traps to EL2
6331:
634
635#ifdef CONFIG_COMPAT
636	msr	hstr_el2, xzr			// Disable CP15 traps to EL2
637#endif
638
639	/* EL2 debug */
640	mrs	x1, id_aa64dfr0_el1		// Check ID_AA64DFR0_EL1 PMUVer
641	sbfx	x0, x1, #8, #4
642	cmp	x0, #1
643	b.lt	4f				// Skip if no PMU present
644	mrs	x0, pmcr_el0			// Disable debug access traps
645	ubfx	x0, x0, #11, #5			// to EL2 and allow access to
6464:
647	csel	x3, xzr, x0, lt			// all PMU counters from EL1
648
649	/* Statistical profiling */
650	ubfx	x0, x1, #32, #4			// Check ID_AA64DFR0_EL1 PMSVer
651	cbz	x0, 6f				// Skip if SPE not present
652	cbnz	x2, 5f				// VHE?
653	mov	x1, #(MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT)
654	orr	x3, x3, x1			// If we don't have VHE, then
655	b	6f				// use EL1&0 translation.
6565:						// For VHE, use EL2 translation
657	orr	x3, x3, #MDCR_EL2_TPMS		// and disable access from EL1
6586:
659	msr	mdcr_el2, x3			// Configure debug traps
660
661	/* Stage-2 translation */
662	msr	vttbr_el2, xzr
663
664	cbz	x2, install_el2_stub
665
666	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
667	isb
668	ret
669
670install_el2_stub:
671	/* Hypervisor stub */
672	adr_l	x0, __hyp_stub_vectors
673	msr	vbar_el2, x0
674
675	/* spsr */
676	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
677		      PSR_MODE_EL1h)
678	msr	spsr_el2, x0
679	msr	elr_el2, lr
680	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
681	eret
682ENDPROC(el2_setup)
683
684/*
685 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
686 * in w0. See arch/arm64/include/asm/virt.h for more info.
687 */
688set_cpu_boot_mode_flag:
689	adr_l	x1, __boot_cpu_mode
690	cmp	w0, #BOOT_CPU_MODE_EL2
691	b.ne	1f
692	add	x1, x1, #4
6931:	str	w0, [x1]			// This CPU has booted in EL1
694	dmb	sy
695	dc	ivac, x1			// Invalidate potentially stale cache line
696	ret
697ENDPROC(set_cpu_boot_mode_flag)
698
699/*
700 * These values are written with the MMU off, but read with the MMU on.
701 * Writers will invalidate the corresponding address, discarding up to a
702 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
703 * sufficient alignment that the CWG doesn't overlap another section.
704 */
705	.pushsection ".mmuoff.data.write", "aw"
706/*
707 * We need to find out the CPU boot mode long after boot, so we need to
708 * store it in a writable variable.
709 *
710 * This is not in .bss, because we set it sufficiently early that the boot-time
711 * zeroing of .bss would clobber it.
712 */
713ENTRY(__boot_cpu_mode)
714	.long	BOOT_CPU_MODE_EL2
715	.long	BOOT_CPU_MODE_EL1
716/*
717 * The booting CPU updates the failed status @__early_cpu_boot_status,
718 * with MMU turned off.
719 */
720ENTRY(__early_cpu_boot_status)
721	.long 	0
722
723	.popsection
724
725	/*
726	 * This provides a "holding pen" for platforms to hold all secondary
727	 * cores are held until we're ready for them to initialise.
728	 */
729ENTRY(secondary_holding_pen)
730	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
731	bl	set_cpu_boot_mode_flag
732	mrs	x0, mpidr_el1
733	mov_q	x1, MPIDR_HWID_BITMASK
734	and	x0, x0, x1
735	adr_l	x3, secondary_holding_pen_release
736pen:	ldr	x4, [x3]
737	cmp	x4, x0
738	b.eq	secondary_startup
739	wfe
740	b	pen
741ENDPROC(secondary_holding_pen)
742
743	/*
744	 * Secondary entry point that jumps straight into the kernel. Only to
745	 * be used where CPUs are brought online dynamically by the kernel.
746	 */
747ENTRY(secondary_entry)
748	bl	el2_setup			// Drop to EL1
749	bl	set_cpu_boot_mode_flag
750	b	secondary_startup
751ENDPROC(secondary_entry)
752
753secondary_startup:
754	/*
755	 * Common entry point for secondary CPUs.
756	 */
757	bl	__cpu_setup			// initialise processor
758	bl	__enable_mmu
759	ldr	x8, =__secondary_switched
760	br	x8
761ENDPROC(secondary_startup)
762
763__secondary_switched:
764	adr_l	x5, vectors
765	msr	vbar_el1, x5
766	isb
767
768	adr_l	x0, secondary_data
769	ldr	x1, [x0, #CPU_BOOT_STACK]	// get secondary_data.stack
770	mov	sp, x1
771	ldr	x2, [x0, #CPU_BOOT_TASK]
772	msr	sp_el0, x2
773	mov	x29, #0
774	b	secondary_start_kernel
775ENDPROC(__secondary_switched)
776
777/*
778 * The booting CPU updates the failed status @__early_cpu_boot_status,
779 * with MMU turned off.
780 *
781 * update_early_cpu_boot_status tmp, status
782 *  - Corrupts tmp1, tmp2
783 *  - Writes 'status' to __early_cpu_boot_status and makes sure
784 *    it is committed to memory.
785 */
786
787	.macro	update_early_cpu_boot_status status, tmp1, tmp2
788	mov	\tmp2, #\status
789	adr_l	\tmp1, __early_cpu_boot_status
790	str	\tmp2, [\tmp1]
791	dmb	sy
792	dc	ivac, \tmp1			// Invalidate potentially stale cache line
793	.endm
794
795/*
796 * Enable the MMU.
797 *
798 *  x0  = SCTLR_EL1 value for turning on the MMU.
799 *
800 * Returns to the caller via x30/lr. This requires the caller to be covered
801 * by the .idmap.text section.
802 *
803 * Checks if the selected granule size is supported by the CPU.
804 * If it isn't, park the CPU
805 */
806ENTRY(__enable_mmu)
807	mrs	x1, ID_AA64MMFR0_EL1
808	ubfx	x2, x1, #ID_AA64MMFR0_TGRAN_SHIFT, 4
809	cmp	x2, #ID_AA64MMFR0_TGRAN_SUPPORTED
810	b.ne	__no_granule_support
811	update_early_cpu_boot_status 0, x1, x2
812	adrp	x1, idmap_pg_dir
813	adrp	x2, swapper_pg_dir
814	msr	ttbr0_el1, x1			// load TTBR0
815	msr	ttbr1_el1, x2			// load TTBR1
816	isb
817	msr	sctlr_el1, x0
818	isb
819	/*
820	 * Invalidate the local I-cache so that any instructions fetched
821	 * speculatively from the PoC are discarded, since they may have
822	 * been dynamically patched at the PoU.
823	 */
824	ic	iallu
825	dsb	nsh
826	isb
827	ret
828ENDPROC(__enable_mmu)
829
830__no_granule_support:
831	/* Indicate that this CPU can't boot and is stuck in the kernel */
832	update_early_cpu_boot_status CPU_STUCK_IN_KERNEL, x1, x2
8331:
834	wfe
835	wfi
836	b	1b
837ENDPROC(__no_granule_support)
838
839#ifdef CONFIG_RELOCATABLE
840__relocate_kernel:
841	/*
842	 * Iterate over each entry in the relocation table, and apply the
843	 * relocations in place.
844	 */
845	ldr	w9, =__rela_offset		// offset to reloc table
846	ldr	w10, =__rela_size		// size of reloc table
847
848	mov_q	x11, KIMAGE_VADDR		// default virtual offset
849	add	x11, x11, x23			// actual virtual offset
850	add	x9, x9, x11			// __va(.rela)
851	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
852
8530:	cmp	x9, x10
854	b.hs	1f
855	ldp	x11, x12, [x9], #24
856	ldr	x13, [x9, #-8]
857	cmp	w12, #R_AARCH64_RELATIVE
858	b.ne	0b
859	add	x13, x13, x23			// relocate
860	str	x13, [x11, x23]
861	b	0b
8621:	ret
863ENDPROC(__relocate_kernel)
864#endif
865
866__primary_switch:
867#ifdef CONFIG_RANDOMIZE_BASE
868	mov	x19, x0				// preserve new SCTLR_EL1 value
869	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
870#endif
871
872	bl	__enable_mmu
873#ifdef CONFIG_RELOCATABLE
874	bl	__relocate_kernel
875#ifdef CONFIG_RANDOMIZE_BASE
876	ldr	x8, =__primary_switched
877	adrp	x0, __PHYS_OFFSET
878	blr	x8
879
880	/*
881	 * If we return here, we have a KASLR displacement in x23 which we need
882	 * to take into account by discarding the current kernel mapping and
883	 * creating a new one.
884	 */
885	msr	sctlr_el1, x20			// disable the MMU
886	isb
887	bl	__create_page_tables		// recreate kernel mapping
888
889	tlbi	vmalle1				// Remove any stale TLB entries
890	dsb	nsh
891
892	msr	sctlr_el1, x19			// re-enable the MMU
893	isb
894	ic	iallu				// flush instructions fetched
895	dsb	nsh				// via old mapping
896	isb
897
898	bl	__relocate_kernel
899#endif
900#endif
901	ldr	x8, =__primary_switched
902	adrp	x0, __PHYS_OFFSET
903	br	x8
904ENDPROC(__primary_switch)
905